IRC logs for #openrisc Saturday, 2013-07-13

--- Log opened Sat Jul 13 00:00:24 2013
poke53281Today I have played a little bit with source code of gcc to implement varargs that behaves like non-varargs.02:24
poke53281The pushing of args onto the stack works, unfortunately the stack pointer r1 does not reckognize the change. *pretend_sizep which should take care of it seems to be useless. It does not change anything. A lot of other architectures implement it like I did.02:25
poke53281Ahh, probably found the problem03:18
stekernjust so I understand what the setup_incoming_varargs function is supposed to do, it takes the registers and put them on the stack, in the same way as if they had been on the stack (i.e. the way it works with the or1k abi), right?03:24
poke53281Yes, the only thing missing is that the stack pointer handles it properly. But this is done in the or1k_expand_prologue somewhere03:25
poke53281or in or1k_compute_frame_size03:26
poke53281So I have to do it more or less manually03:26
stekernyeah, I think *you* have to use that *pretend_sizep information somewhere03:27
poke53281Exactly. I have to include crtl->args.pretend_args_size03:27
poke53281somewhere in the function03:28
stekernmore precisely: or1k_compute_framesize: save it into the frame_info struct and allocate stack space. or1k_expand_prologue: put the varargs regs on the stack *first*, update stack pointer. or1k_expand_epilogue: update stack pointer03:42
poke53281Yes, writing it in an english sentence is easier than actually doing it :)03:48
poke53281what is the redzone or or1k?04:00
stekern128 bytes04:05
stekernhmm, a lot of the c++ tests are failing...04:18
stekernpoke53281: ping me if you intend to run the testsuite in the near future04:19
poke53281The C-tests are fine?04:20
stekernI'd like to know if it's changes in 4.8.1 that's causing this or if it's the same in 4.8.004:20
stekernthere are a couple of fails there too, but I think those have been there earlier04:21
poke53281well, take the first one which is failing and check if this is also present in 4.8.004:21
poke53281I think I can manage it tomorrow04:23
stekernI'm waiting for those tests to finish first04:23
stekern...then I need to reboot that machine and invetigate why my graphics card suddenly broke down...04:24
poke53281You compiled your graphics driver accidentally for the or1k- platform?04:26
stekernI would have been surprised if it would have worked as long as it did then04:30
stekernI think at least one test fails after fixing those constructors/destructors, but I think that have been a false positive before05:35
poke53281I have bad news. The computer on which I am working doesn't have a tap device. And I don't have root privileges. So connection with or1ksim will not work.05:53
poke53281Have to switch the computer.05:54
stekernlooks like a lot of failures are of the type: throw exception -> segfault09:12
stekernat least the libstdc++ test works now, they didn't in 4.8.013:23
stekernohh, it looks like sjlj exceptions are not used, but they should be I think13:36
stekernblueCmd: you around?13:37
stekerndoh, I haven't copied my to the right place in my build root...14:21
poke53281Ok :)17:00
poke53281So, I have installed everything on another computer with tap device. Let's see17:01
stekernpoke53281: good luck! ;)17:16
stekerntechnically, you could have just installed or1ksim on that, as long as you have net access to that box17:17
stekernI'm rerunning the tests with the right libgcc_s now... at least a handful of the test I checked by hand passed with that17:18
stekernI think blueCmd changed so we're not using sjlj anymore17:19
olofkFYA, I used the "silent ack" card and applied some fixes to or1200-monitor in orpsocv219:11
olofkI think we should set up a new teleconference pretty soon. It would be good to have some discussions about orconf 201319:12
olofkMaybe we should try something else than hangouts then, as there seemed to be more than ten people who wanted to join19:13
jeremybennettolofk: Yes. Simple teleconf may be easiest. Let me know if you want me to set it up.19:14
olofkThe best feature with hangout is that we can get a recording of the talk, but we could just use any service and ask NSA for a dump of the session if we need it19:14
olofkjeremybennett: That would be great. I'm a bit too old school for setting up those kind of things19:15
poke53281stekern: testsuite running. Currently without the _init/_fin changes, but with the gcc-native patch19:20
poke53281How long will this run?19:21
olofkstekern or poke53281: It would be awesome if someone could put up a guide for running the GCC testsuite on the wiki. I've been interested in doing that a few times, but never had the time to read up on how to do it19:24
jeremybennettolofk: OK - give me a day or two to set it up. We can get it recorded as well.19:25
olofkjeremybennett: Perfect. Now we only need some attendants :)19:27
poke53281stekern: I was wrong. pulled yesterday the updates. So _init and _fin are already corrected19:29
blueCmdstekern: correct21:07
blueCmddw2 is used noq21:08
olofkhmm... for some reason, or1200-monitor doesn't print out (exception) anymore22:23
olofkOr rather, the reason is that wb_freeze and except_flushpipe aren't asserted at the same time. Has there been any changes in that area recently?22:25
olofksorry.. wb_freeze should _not_ be asserted22:26
olofkoh... running a testcase that actually triggers exception seems to help23:48
--- Log closed Sun Jul 14 00:00:26 2013

Generated by 2.15.2 by Marius Gedminas - find it at!